OpticalFlow-RI is a source code repository of tested and validated optical flow algorithm implementations, being specificaly tuned for fluid mechanics applications with Particle Imaging Velocimetry (PIV) image databases having a maximum displacement of 4.0px between frames.
This repository intends to be a reference library for reuse by anyone interested in evaluating said algorithms for fluid mechanics velocimetry applications.
The library was created with minimal third-party dependencies for the relevant algorithms. This way it is less susceptible to unwanted behavioral changes that may arise by third-party libraries implementation changes/upgrades, ultimately leading to uncalibrated results or API breakage.
The library includes concrete examples on how to run the algorithms along with the calibration parameters.
Current reference implementations include: Horn-Schunck (Python, Numba), Lucas-Kanade (Python, OpenCL), Farnebäck (Python, OpenCL) and Liu-Shen Physics based optimizer (Python, Numba)
Open-Source licensed
| 
|
Python3 is required to run these algorithms.
Implementations tested with Ubuntu 20.04.2LTS and Ubuntu 21.04
Required packages include: apt install python3 python3-numpy python3-skimage python3-pyopencl python3-numba
Note: A valid OpenCL device is required to run dense Lucas-Kanade and Farnebäck algorithms.
There are a number of examples in the examples folder, namely:
| Example file | Description | Execute |
|---|---|---|
| Farneback_Fs0_0.py | Farnebäck with Null Gaussian pre-filter without pyramidal | python3 Farneback_Fs0_0.py |
| Farneback_Fs0_0_PyrLvls2.py | Farnebäck with Null Gaussian pre-filter with pyramidal | python3 Farneback_Fs0_0_PyrLvls2.py |
| LiuSE_Farneback_Fs0_0_PyrLvls2.py | Liu-Shen optimized Farnebäck with Null Gaussian pre-filter with pyramidal | python3 LiuSE_Farneback_Fs0_0_PyrLvls2.py |
| denseLK_Fs2_0.py | dense Lucas-Kanade with 2.0px Gaussian pre-filter without pyramidal | python3 denseLK_Fs2_0.py |
| denseLK_Fs2_0_PyrLvls2.py | dense Lucas-Kanade with 2.0px Gaussian pre-filter with pyramidal | python3 denseLK_Fs2_0_PyrLvls2.py |
| LiuSE_denseLK_Fs2_0_PyrLvls2.py | Liu-Shen optimized dense Lucas-Kanade with 2.0px Gaussian pre-filter with pyramidal | python3 LiuSE_denseLK_Fs2_0_PyrLvls2.py |
| PyHSchunck_Fs3_4.py | Horn-Schunck with 3.4px Gaussian pre-filter without pyramidal | python3 PyHSchunck_Fs3_4.py |
| PyHSchunck_Fs3_4_PyrLvls2.py | Horn-Schunck with 3.4px Gaussian pre-filter with pyramidal | python3 PyHSchunck_Fs3_4_PyrLvls2.py |
| LiuSE_PyHSchunck_Fs3_4_PyrLvls2.py | Liu-Shen optimized Horn-Schunck with 3.4px Gaussian pre-filter with pyramidal | python3 LiuSE_PyHSchunck_Fs3_4_PyrLvls2.py |
The official source repository for Optical Flow reference implementations is located in the CoreRasurae Github repository and can be cloned using the following command.
git clone https://github.com/CoreRasurae/OpticalFlow-RI